Preliminary Study On Discrimination Of Soil Types From Their Visible-near Infrared Spectra

Hyperspectral technology has been widely applied to prediction of the soilproperty(soil organic matter, soil water, iron oxide, etc.) and soil salinization, soilerosion, soil pollution research in recent years. Soil is the basis of agriculturalproduction and the important carrier of human survival environment, sodiscrimination of soil types based on hyperspectral data is of great significance foragricultural production and management. In this paper, using the Netherlands Avantescompany Avaspec-2048x14, Avaspec-NIR256-2.5two features spectrometermeasured visible light and near infrared spectrum curve of25soil profiles. Using thelogarithm of inverse, first derivative, remove the envelope three spectral datatransformation methods extracted spectral characteristics of the topsoil and subsoil ofsoil profiles. Using principal component analysis method to extract the principalcomponent clustered and distinguished of different kinds of soil. The paper discussedfeasibility of discrimination of the different soil types based on hyperspectralreflectance characteristics. Given acquisition methods of soil spectra test data werenot unified, in order to discrimination of different soil types better, soil sample surfacetreatment way and particle size that effected on the stability of high spectrum testingsoil sample were studied through analysis of variance.The specific content andconclusions are as follows:(1) In indoor soil hyperspectral test, different methods of soil sample surfacetreatment and soil sample particle size affected the result of the test. The soilreflectance of soil sample surface planishing is bigger than soil sample surface sicking;Soil spectral reflectance as test soil samples increases with the decrease of particlesize.(2) Soil sample surface treatment method and the soil sample particle size hadsignificant effect on the stability of high spectrum test result. Under different ways ofsoil sample surface treatment, sicking was better than planishing. Under different testparticle size, soil particle of medium size was better and the soil particle of1mm wasthe most stable on sicking, soil particle of small size is better than soil particle of bigsize on planishing. In unified indoor geometry test conditions, the sicking and1mmsoil particle could obtain relatively stable soil hyperspectral data. (3) In spectral data transformation processing, the logarithm of inverse, firstderivative, continuum removal three transform method could effectively extractspectral curve features and enhance spectral information. Soil profiles spectral curvefeatures was similar on the whole, the position and the relative strength of absorptioncharacteristics of different types of soil was different. Overall, absorptioncharacteristics in visible light was less, absorption characteristics were mainlydistributed in near infrared wave band. There were6absorption characteristics of430nm-550nm,800nm-990nm,1300nm-1500nm,1600nm-1700nm,1800nm-2100nm,2150nm-2300nm.(4) Principal component analysis of based on topsoil spectrum and based on thetopsoil and subsoil spectrum could distinguish between soil orders, the latter is muchbetter than the former. Discrimination results of soil hyperspectral data treated bydifferent transformation methods were different, the continuum removal method wasthe best, the logarithm of inverse was better than first derivative.